Can You Rely on AI for Accounting and Tax Advice?

AI is becoming part of everyday business life.

I use it myself, and I think it can be extremely useful.

But over the last couple of months, we’ve had several examples where clients have taken accounts, accounting queries or tax advice we’ve sent them, run them through AI, and then come back with additional questions.

Some of those questions have been useful.

Some have been completely unnecessary.

And some have simply been wrong.

That’s the problem with using AI for accounting and tax without understanding the subject you’re asking about.

AI Can Sound Confident Even When It’s Wrong

One of the biggest risks with AI is that it can produce an answer that sounds very convincing.

The wording may be polished.

The explanation may sound technical.

It may even quote rules, percentages or accounting terminology.

But that doesn’t necessarily mean the answer is correct.

AI can sometimes hallucinate — in other words, produce information that sounds plausible but isn’t actually supported by the facts or the relevant rules.

That becomes dangerous when the person reading the answer doesn’t have enough knowledge of the subject to recognise that something is wrong.

When AI Creates Questions That Don’t Really Need Asking

We recently sent a client their final accounts.

The client then ran the information through AI and came back asking for a breakdown of various balance sheet figures.

The difficulty was that the information had effectively already been provided.

The real issue wasn’t that the accounts were missing information.

It was that the client didn’t fully understand what had been sent to them.

AI had generated a set of further questions, but those questions didn’t necessarily add anything useful.

This is an important distinction.

AI is very good at generating questions.

But that doesn’t automatically mean those are the right questions.

Some Accounting Questions Need to Be Answered by the Business

We’ve also had cases where clients have apparently put our year-end accounting queries through AI and then sent back additional questions.

Some of those questions related to areas such as the recoverability of trade debtors.

But that is often something the directors or management of the business need to assess.

An accountant can ask the question.

We can challenge the answer.

We can look at the age of the debt, subsequent receipts and the evidence available.

But management ultimately has information about customers and commercial circumstances that an external AI tool simply doesn’t have.

The same applies to many year-end accounting judgements.

AI cannot replace the information and judgement that exists inside the business.

Tax Advice Is Even More Dangerous

Tax is another area where people need to be particularly careful.

We’ve provided tax advice to clients who have then apparently put that advice through AI and come back with further questions.

Again, there is nothing wrong with doing that.

In fact, asking questions is a good thing.

But some of the follow-up questions we’ve seen have been based on incorrect assumptions or information that simply wasn’t relevant to the situation.

Tax rules are detailed.

The answer can depend on the year involved, the type of taxpayer, the nature of the transaction and many other facts.

If AI starts with the wrong assumption, it can take you down completely the wrong path.

And if you don’t understand the subject well enough, you may not realise that’s happening.

Always Ask: Where Has This Information Come From?

This is one of the most important habits when using AI.

If it gives you a tax or accounting answer, ask:

What is the source?

For UK tax, I’d want to know whether the answer is based on:

• HMRC guidance
• GOV.UK
• legislation
• accounting standards
• another reliable technical source

You should also ask AI to explain why it has reached its conclusion.

If it cannot point you towards a credible source, that should immediately make you more cautious.

You Need Some Understanding of the Subject

This is probably the main point.

AI works much better when the person using it already understands at least something about the subject.

If you know the basics, you can recognise when an answer looks odd.

You can challenge assumptions.

You can ask better follow-up questions.

And you can tell the difference between a useful point and complete nonsense.

If you know nothing about the subject, you are much more vulnerable to simply accepting whatever AI tells you.

That applies to accounting, tax, law, finance and probably many other areas.

AI Is a Tool, Not a Substitute for Judgement

I’m certainly not anti-AI.

I use it a lot.

It can be excellent for:

• summarising information
• generating ideas
• explaining technical concepts
• reviewing wording
• identifying questions you may want to ask
• helping you research a subject

But it still needs judgement.

It doesn’t know your business in the way you do.

It doesn’t necessarily know all the facts.

And it may not understand the context behind an accounting or tax issue.

That is why professional judgement still matters.

The Better Way to Use AI

If you want to use AI to review accounting or tax information, I’d suggest a simple approach.

First, give it the full context.

Second, ask it to explain the reasoning behind its answer.

Third, ask it to provide reliable sources.

Fourth, challenge anything that doesn’t look right.

And finally, if the issue is important, check it with somebody who understands the subject.

AI can help you ask better questions.

It shouldn’t automatically be treated as the final answer.

The Bottom Line

AI is useful.

I use it myself, and I think businesses will increasingly use it to review accounts, tax advice and financial information.

But there is a danger in treating a confident AI response as though it must be correct.

If you’re using AI for accounting or tax, you need enough understanding of the subject to challenge what it tells you.

And if the answer matters, ask for the source.

Use AI as a tool to support your judgement — not as a replacement for it.